The people who build your evidence have sat on the other side of the table.
Dr Nikhil KhadabadiChief Medical Officer, Orthopedics and SpineFormer TÜV SÜD senior reviewer
Dr Mark Da CostaChief Operating Officer and Head of CardiovascularFormer TÜV SÜD Team Leader
Pierre-Marie BoutanquoiHead of Medical WritingCER, PMCF plans and reports
Registry or clinical investigation? Which route closes the open question on your device, what each one can really prove, and the eleven questions we work through before recommending either.
Pick a registry when the open question is how an established device behaves in routine care, over time, across a wide population. Pick a clinical investigation when the question needs conditions routine care does not provide: a defined population, a predefined endpoint, a controlled assessment, sometimes a comparator. Everything below is the reasoning behind that sentence, plus the eleven questions we work through before we recommend one route or the other.
A registry and a clinical investigation are not always separate categories. A registry describes how data are collected, while "clinical investigation" is a regulatory classification based on the study purpose and design. Some prospective registries may therefore be conducted as clinical investigations.
Almost nobody arrives with a clean question. They arrive with a Notified Body comment, a PMCF plan that is due for an update, or a claim the commercial team wants and the clinical file does not yet carry.
The question asked out loud is usually the wrong one: what study should we run. The useful question comes earlier. What do we still not know about this device, and what is the smallest piece of evidence that closes that gap. Four routes tend to sit on the table at the same time.
Continuous, structured data collection on patients treated in routine clinical practice.
Protocol driven, narrower than a registry, usually a fixed sample and a fixed follow up window.
A planned study with predefined endpoints, and the regulatory and ethics pathway that comes with it.
Two of the above, sequenced, written as one evidence plan.
They get confused because they overlap in the middle. All of them can feed post-market clinical follow-up under Annex XIV Part B of Regulation (EU) 2017/745. All of them produce clinical data. All of them end up in the clinical evaluation report. The difference is not the label on the study. It is how much control you have over what gets measured, on whom, and when.
One thing to be clear about before you read on. This page is about the choice between the two routes. It is not a guide to running a registry, and it is not a PMCF strategy. If you already know you need a registry, the delivery side sits on our registry under MDR page, and the wider picture sits on the PMCF hub.
A registry collects structured data on patients who receive a device or a procedure in routine clinical practice. There is no protocol driven intervention. Care happens the way it would have happened anyway, and the registry records it in a consistent format so the data can be pooled and analyzed.
Registries come in several shapes, and the shape matters far more than the word.
The data model is fixed first, then patients are enrolled and followed forward. You decide what is captured, so completeness and traceability are usually much better.
Data that already exist are pulled and structured after the fact. Fast and inexpensive, but you inherit whatever was recorded and, more to the point, whatever was not.
The manufacturer owns the protocol, the data model and the dataset. Full control over the variables, full cost, full responsibility for sites and data quality.
An arthroplasty registry, a cardiac registry, a national procedure registry. Large populations, long follow up, real benchmarks. You take the data model as it is, and you negotiate access and data use.
A registry joined to another source, for example an administrative or claims dataset, to extend follow up or capture outcomes the registry itself does not record.
Not prospective against retrospective. If you control the data model, a registry can answer a much narrower question than people expect. If you do not, it answers whatever question its own designers had in mind, and you work with that.
Where the registry route is chosen to generate post-market evidence at scale, it sits inside the wider real-world evidence discipline.
A registry does not change how patients are treated. It changes whether what happened can be read back.
A clinical investigation is a planned study designed to answer predefined questions about the safety, the performance or the clinical benefit of a device. The protocol comes first. Population, endpoints, visit schedule and analysis are all fixed before the first patient is enrolled.
That Article 74 line, additional burdensome procedures or not, is very often the real border between a registry style PMCF activity and a clinical investigation. It is worth checking early, because it changes the budget and the timeline more than any other single decision on this page. The design side is covered on our page on clinical investigations with medical devices in Europe, and the standard itself in our ISO 14155 training.
The short version, written for a device already on the market or close to it, which is where the question usually comes up.
| Dimension | Registry | Clinical investigation |
|---|---|---|
| Primary purpose | Describe how a device performs in routine care, over time and at scale. | Answer a predefined question about safety, performance or clinical benefit. |
| Level of control | Variable. Low for an external registry, potentially high for a sponsor-owned prospective registry. | High. The protocol fixes what happens and when. |
| Patient selection | Consecutive or all comers. Broad, and close to the real treated population. | Set by inclusion and exclusion criteria. Narrower and cleaner. |
| Intervention | None beyond routine care. | May include procedures, assessments or visits that would not otherwise happen. |
| Endpoint definition | Existing registries inherit established variables and definitions. Sponsor-owned registries can prespecify endpoints and follow-up. | Predefined, with a definition, a timepoint and an analysis written in advance. |
| Data completeness | Variable and dependent on registry design, site processes and active follow-up. | Managed. Edit checks, queries and active follow up on missing data. |
| Monitoring | May range from centralized quality checks to risk-based site monitoring. | Set out in a monitoring plan, on site or remote, per ISO 14155. |
| Source-data verification | Limited or none, particularly in external and national registries. | Planned, with source documents available for the monitored variables. |
| Follow-up | Long, sometimes many years, but it depends on routine visits actually happening. | Fixed schedule and fixed window. Long follow up is possible and expensive. |
| Regulatory pathway | Depends on design and country. Ethics and data protection approvals are still required. | Article 62 and Annex XV before CE marking, Article 74 for PMCF investigations, ISO 14155 as the GCP standard. |
| Cost and timeline | Lower per patient, longer horizon. Cost driven by duration and data management. | Higher per patient, faster to a defined answer. Cost driven by sites, visits and monitoring. |
| Strengths and limitations | Strong on scale, real world relevance and long term signals. Weak on confounding, missing data and control over what is measured. | Strong on internal validity and on defending one specific answer. Weak on population breadth, scale and speed to start. |
Read the table as a trade. You give up control to gain scale, or you give up scale to gain control. There is no route that gives you both.
A registry earns its place when the question is about duration, volume or ordinary practice.
A clinical investigation becomes the right answer, and sometimes the only answer, when routine practice cannot produce the data.
Look at the actual variables in the actual registry, not at the registry's reputation. If the endpoint you need is recorded nowhere, no amount of sample size fixes it. That single check settles more route decisions than any other.
This is the framework we use in a first working session. Two or three of these usually settle it on their own. Question 3 and question 7 do most of the work.
Before CE marking you are generally in Article 62 territory and a registry is not a substitute. After CE marking both routes open up, and Article 74 decides which side of the line a post-market study sits on.
Established technology with a long clinical history can often be described from routine data. Novel technology has no routine data yet, and nothing sensible to compare against.
Write it as one sentence. If it will not fit in one sentence, no study design is going to save it. That sentence gives you the endpoint, and the endpoint gives you the route.
A new claim needs evidence collected against that claim. Registries describe what happened. They do not usually confirm what you want to say.
Look at the actual variables in the actual registry, not at the registry's reputation. If the variable is not in the dataset, the answer is no, and no amount of sample size fixes it.
If you cannot tell which patient received your device, and which version of it, registry data will not support a device-specific conclusion. UDI capture is the practical test.
The sharper version of the question above. Endpoint definition, timepoint, and who adjudicates it. All three, not just the first.
Routine care produces imaging at variable timepoints, read locally, under variable protocols. If you need a core lab read or a validated PROM at a fixed timepoint, you are adding procedures, and you have moved toward the investigation side.
A registry can give you a benchmark, national or internal. It rarely gives you a controlled comparison. If you need control of confounding, you need a design that provides it.
This is the Article 74 line, and it is the one that changes the budget. Additional burdensome procedures move you into the clinical investigation regime.
Different audiences accept different evidence. Payers often accept real-world data that a Notified Body would push back on for a performance claim. Decide the audience before you decide the design.
The decision fork. Question 3 names the uncertainty, question 7 tells you whether routine data already carry the endpoint.
Registries are useful and they are not free of problems. These are the ones that actually cost projects time.
The other side of the trade, and it is a real one.
In practice the strongest pathway is often not one route or the other. It is both, sequenced, and written as a single evidence plan.
For the early or device-specific question. Small, tight, designed to close one named gap and nothing more.
It picks up where the investigation stops, at a much lower cost per patient year.
A defined add-on module can reach endpoints a plain registry cannot. Keep one eye on the Article 74 line while you design it.
The retrospective part sizes the signal and justifies the design. The prospective part answers the question.
The condition for any of this working is that the endpoint definitions match across the two pieces. If they do not, you end up with two datasets that cannot be read together, which is the most expensive mistake on this page.
Whichever route you take, the evidence has to land somewhere, and it lands in the same place.
A registry usually reports into the PMCF Evaluation Report, in the structure of MDCG 2020-8, and feeds the clinical evaluation as real-world evidence. A clinical investigation reports as a clinical study report and feeds the clinical evaluation as study data with predefined endpoints. The PMCF plan itself, in the structure of MDCG 2020-7, is where the choice gets documented and justified in the first place.
Which is the practical reason to make this decision early. The route determines what the CER can conclude, and the CER is what a Notified Body reads.
We work on both routes, and most of our work happens before either one starts.
Naming the residual uncertainty in one sentence, which is harder than it sounds.
Working the eleven questions above with your regulatory and clinical teams.
Endpoints that answer the question and survive review.
Article 62, Article 74, national requirements, ethics and data protection.
Sites, volumes, data model, custodian access, and whether the variables you need are really there.
Full-service conduct under ISO 14155, sites, monitoring and project management.
EDC, data quality, analysis plan and reporting.
PMCF plans and PMCF Evaluation Reports.
Getting the output into the clinical evaluation so it is usable, not just filed.
Sometimes, and it depends entirely on the question. If the open question is long-term survivorship, revision rate or rare complications on an established device, a well designed registry is often the better instrument, not just the cheaper one. If the question is whether a new design performs as claimed against a defined endpoint, a registry will not close it, because the data you need were never collected in routine care. The test is not the size of the dataset. It is whether the dataset contains the variable that answers the question.
Yes, and it happens more often than people expect. ISO 14155:2026 applies to clinical investigations of medical devices, and a sponsor-owned prospective registry can be run under a clinical investigation plan with ISO 14155 discipline: written protocol, defined endpoints, monitoring plan, source data. Doing that raises cost and site burden considerably. It is worth it when the evidence has to support a claim or answer a Notified Body, and it is overkill when the registry exists to describe routine performance.
A registry does not become a clinical investigation simply because an additional data field, questionnaire or follow-up contact is introduced. Classification depends on the study purpose and design. For a CE-marked device used within its intended purpose, Article 74(1) is particularly relevant when procedures are additional to normal use and are invasive or burdensome. Use outside the intended purpose falls under Article 74(2). National ethics and competent-authority requirements should also be confirmed.
It can, as a component. Retrospective data are useful for describing what has already happened, sizing a signal, and justifying the design of whatever comes next. What they rarely do on their own is close a residual uncertainty, because they were not collected to answer your question. The pattern that works is retrospective first to establish the baseline, prospective follow up to answer the specific point, and both written into the PMCF plan as one activity rather than two.
Rarely on its own. A new indication is a new claim, and a new claim needs evidence collected against a defined endpoint in the population you are claiming for. Registry data can support the case, for example by showing that the population exists and that use in it is already happening without a safety signal. The confirmatory piece usually still has to be a clinical investigation. Expect the registry to frame the question and the investigation to answer it.
Then it does not answer your question, whatever its size. You have three realistic options: add the missing variables if it is your own registry, negotiate a module or a sub-study if it is an external registry and the custodian allows it, or run a focused clinical investigation alongside for the endpoint you actually need. What does not work is analyzing a proxy variable and hoping it stands in. Reviewers read that for what it is.
Yes, and it is often the strongest strategy available. The investigation carries the specific, device-level, endpoint-driven question on a defined population over a defined window. The registry carries scale and duration, which is where implant survivorship and rare events actually appear. Running both means one evidence plan, not two, so the endpoints and the definitions have to be aligned from the start or the two datasets will never be readable together.
A registry is almost always cheaper per patient and slower to a definitive answer. A clinical investigation costs more per patient and gets you a clean answer sooner, because the endpoint was designed to be answerable. So the honest version of the question is not which is cheaper, it is what a delay costs you. If a Notified Body deficiency is holding up a certificate, the fast expensive route is the cheap one.
It changes what the clinical evaluation report can conclude, and how strongly. Registry output feeds the CER as real-world evidence: good for safety, longevity and use in broad populations, weaker on causal claims. Clinical investigation output feeds it as a clinical study report with predefined endpoints, which is what performance and benefit claims need. Both belong in the CER. Just do not expect one to carry weight it was never designed to carry.
On relevance and quality, not on volume. The questions that come back are consistent: can the device be identified at individual patient level, is the population the one covered by the intended purpose, are the endpoints defined and are they the ones the claim needs, how much data are missing and how was that handled, and is the follow up long enough to see the event of interest. Registry evidence that answers those holds up. Registry evidence submitted mainly because it happened to be available does not.
Eclevar manages RegenLab's PMCF programme on chronic wound devices. This is a randomized study of 160 patients across 14 sites in 5 EU countries, covering both diabetic foot ulcer (DFU) and venous leg ulcer (VLU) indications. The partnership combines Eclevar's ISO 14155 clinical expertise with the Milo Studio platform to deliver post-market clinical follow-up evidence that supports both Notified Body scrutiny and reimbursement endpoints, from protocol design through to final study report.
The pages that carry the delivery detail this one deliberately leaves out.
Send us the device, the intended purpose and the open question, and we will come back with a recommended route and the reasoning behind it. The entry point is a 45 minute evidence-route review: you send the intended purpose, the current clinical evaluation conclusion and the open point, and you get back a written recommendation, registry or clinical investigation or both, with the eleven questions answered for your device.
Sometimes, and it depends entirely on the question. If the open question is long-term survivorship, revision rate or rare complications on an established device, a well designed registry is often the better instrument, not just the cheaper one. If the question is whether a new design performs as claimed against a defined endpoint, a registry will not close it, because the data you need were never collected in routine care. The test is not the size of the dataset. It is whether the dataset contains the variable that answers the question.
Yes, and it happens more often than people expect. ISO 14155:2026 applies to clinical investigations of medical devices, and a sponsor-owned prospective registry can be run under a clinical investigation plan with ISO 14155 discipline: written protocol, defined endpoints, monitoring plan, source data. Doing that raises cost and site burden considerably. It is worth it when the evidence has to support a claim or answer a Notified Body, and it is overkill when the registry exists to describe routine performance.
A registry does not become a clinical investigation simply because an additional data field, questionnaire or follow-up contact is introduced. Classification depends on the study purpose and design. For a CE-marked device used within its intended purpose, Article 74(1) is particularly relevant when procedures are additional to normal use and are invasive or burdensome. Use outside the intended purpose falls under Article 74(2). National ethics and competent-authority requirements should also be confirmed.
It can, as a component. Retrospective data are useful for describing what has already happened, sizing a signal, and justifying the design of whatever comes next. What they rarely do on their own is close a residual uncertainty, because they were not collected to answer your question. The pattern that works is retrospective first to establish the baseline, prospective follow up to answer the specific point, and both written into the PMCF plan as one activity rather than two.
Rarely on its own. A new indication is a new claim, and a new claim needs evidence collected against a defined endpoint in the population you are claiming for. Registry data can support the case, for example by showing that the population exists and that use in it is already happening without a safety signal. The confirmatory piece usually still has to be a clinical investigation. Expect the registry to frame the question and the investigation to answer it.
Then it does not answer your question, whatever its size. You have three realistic options: add the missing variables if it is your own registry, negotiate a module or a sub-study if it is an external registry and the custodian allows it, or run a focused clinical investigation alongside for the endpoint you actually need. What does not work is analyzing a proxy variable and hoping it stands in. Reviewers read that for what it is.
Yes, and it is often the strongest strategy available. The investigation carries the specific, device-level, endpoint-driven question on a defined population over a defined window. The registry carries scale and duration, which is where implant survivorship and rare events actually appear. Running both means one evidence plan, not two, so the endpoints and the definitions have to be aligned from the start or the two datasets will never be readable together.
A registry is almost always cheaper per patient and slower to a definitive answer. A clinical investigation costs more per patient and gets you a clean answer sooner, because the endpoint was designed to be answerable. So the honest version of the question is not which is cheaper, it is what a delay costs you. If a Notified Body deficiency is holding up a certificate, the fast expensive route is the cheap one.
It changes what the clinical evaluation report can conclude, and how strongly. Registry output feeds the CER as real-world evidence: good for safety, longevity and use in broad populations, weaker on causal claims. Clinical investigation output feeds it as a clinical study report with predefined endpoints, which is what performance and benefit claims need. Both belong in the CER. Just do not expect one to carry weight it was never designed to carry.
On relevance and quality, not on volume. The questions that come back are consistent: can the device be identified at individual patient level, is the population the one covered by the intended purpose, are the endpoints defined and are they the ones the claim needs, how much data are missing and how was that handled, and is the follow up long enough to see the event of interest. Registry evidence that answers those holds up. Registry evidence submitted mainly because it happened to be available does not.
Whitepapers and publications produced by our teams with our notified body partners.
Written with Notified Body BSI: a practical reading of the clinical evidence expectations under EU MDR 2017/745, the same evidence your file has to support.
The services and topics connected to this page.
Send us the device, the claim and the deadline. You get a written answer within 24 hours.
Your documents are reviewed confidentially. An NDA can be signed before we receive any technical or clinical information.
